Using Large Language Models and Knowledge Graphs to Improve the Interpretability of ML Models

Forum topic · 小凯 · 2026-04-21

Summary

Researchers Thomas Bayer, Alexander Lohr, Sarah Weiß, Bernd Michelberger, and Wolfram Höpken (arXiv:2604.16280) propose a method combining Knowledge Graphs (KG) with Large Language Models (LLM) to improve the explainability of Machine Learning results. Domain-specific data, ML results, and their explanations are stored in a knowledge graph, creating a structured link between domain knowledge and ML insights. A selective retrieval mechanism extracts relevant triples from the KG, which an LLM then converts into user-friendly natural-language explanations. The approach was evaluated in a manufacturing setting using the XAI Question Bank, including custom complex questions beyond standard ones. Across 33 questions, answers were analyzed with quantitative metrics (accuracy, consistency) and qualitative metrics (clarity, usefulness). The work contributes both a novel method for letting LLMs dynamically access KG content for XAI and empirical evidence that such explanations work in real manufacturing environments, supporting better decision-making in manufacturing processes.

Overview

Field: Machine Learning (XAI)
Authors: Thomas Bayer, Alexander Lohr, Sarah Weiß, Bernd Michelberger, Wolfram Höpken
Published: 2026-04-17
arXiv: 2604.16280

Abstract

Explaining Machine Learning (ML) results in a transparent and user-friendly manner remains a challenging task of Explainable Artificial Intelligence (XAI). In this paper, the authors present a method to enhance the interpretability of ML models by using a Knowledge Graph (KG). Domain-specific data is stored along with ML results and their corresponding explanations, establishing a structured connection between domain knowledge and ML insights. To make these insights accessible to users, a selective retrieval method was designed in which relevant triplets are extracted from the KG and processed by a Large Language Model (LLM) to generate user-friendly explanations of ML results.

Evaluation

The method was evaluated in a manufacturing environment using the XAI Question Bank. Beyond standard questions, the authors introduced more complex custom questions to highlight the advantages of their approach. In total, 33 questions were evaluated, with answers analyzed using quantitative metrics (accuracy, consistency) and qualitative metrics (clarity, usefulness).

Contributions

Theoretical: A novel method that effectively enables LLMs to dynamically access a knowledge graph to improve the interpretability of ML results.
Practical: Empirical evidence that such explanations can be successfully applied in a real manufacturing environment, supporting better decision-making in manufacturing processes.

---

*Auto-collected on 2026-04-21*